以兩段式多特徵取向建構鈔券辨識系統之研究

本研究利用數位影像處理技術建構一個新臺幣真偽鈔卷的辨識系統，針對中針民國九十年七月式發行流通之新臺幣各幣值之鈔卷進行研究，提出一個兩段式多特徵的辨識系統（Two-staged Multi-feature Recognition System, TMRS），此系統將鈔卷辯識過程分為兩個步驟來進行：一是粗分類，排除明顯的非鈔卷影像，二是細分類，判斷鈔卷影像是否具備特定的細部防偽特徵。
我們運用快速傅立葉轉換（Fast Fourier Transform, FFT）與Sobel邊界偵測（Sobel Edge Detection）擷取用以鑑識真偽鈔穩定而有效的特徵，並以支持向量機（Support Vector Machine，SVM）作為辨識系統之分類核心，我們進行了一系例的實驗，實驗結悲顯示，對於收收集到之鈔卷影像而言，TMRS具有相當良好的辨識效果。

In recent years, making counterfeit money has gradually become a common criminal problem.
Every country’s bank notes made by very precise and complicated designed. One bank note was used many kinds characteristics to prevent counterfeit. The characteristics was be used universality, for example: watermakeing, specific materials et. Its not only could exhibited the country’s appearance but also convenient for people to make up.
In this paper, we proposed Two-stage Multi-feature Approach to recognition bank notes. This system can differentiate the counterfeit bank notes make by color copy machines and color printer. The experimental results show the proposed method has the best result and easily applied to the computer to get good effects.